ARTICLE ABSTRACTSolid cancer cells escape the primary tumour mass by transitioning from an epithelial-like state to an invasive migratory state. As they escape, metastatic cancer cells employ interchangeable modes of invasion, transitioning between fibroblast-like mesenchymal movement to amoeboid migration, where cells display a rounded morphology and navigate the extracellular matrix in a protease-independent manner. However, the gene transcripts that orchestrate the switch between epithelial, mesenchymal and amoeboid states remain incompletely mapped, mainly due to a lack of methodologies that allow the direct comparison of the transcriptomes of spontaneously invasive cancer cells in distinct migratory states. Here, we report a novel single cell isolation technique that provides detailed 3D data on melanoma growth and invasion, and enables the isolation of live, spontaneously invasive cancer cells with distinct morphologies and invasion parameters. Via the expression of a photoconvertible fluorescent protein, compact epithelial-like cells at the periphery of a melanoma mass, elongated cells in the process of leaving the mass, and rounded amoeboid cells invading away from the mass were tagged, isolated and subjected to single cell RNA-sequencing. 462 differentially expressed genes were identified, from which two candidate proteins were selected for further pharmacological perturbation, yielding striking effects on tumour escape and invasion, in line with the predictions from the transcriptomics data. This work describes a novel, adaptable and readily implementable method for the analysis of the earliest phases of tumour escape and metastasis, and its application to the identification of genes underpinning the invasiveness of malignant melanoma.